Table 4 Results in the holdout set of all models after training on the full training set, ranked by AUCPR

XGBoost (Bayesian Optimization)

0.9974 (0.9961–0.9984)

0.9997 (0.9996–0.9999)

0.995 (0.993–0.9965)

0.9787 (0.9719–0.9849)

0.9586 (0.943–0.9742)

0.9987 (0.9978–0.9996)

0.0041 (0.0031–0.0051)

Random Forest (Bayesian Optimization)

0.9973 (0.996–0.9983)

0.9997 (0.9996–0.9999)

0.9947 (0.9927–0.9963)

0.9749 (0.9676–0.9817)

0.9506 (0.9337–0.9676)

0.9992 (0.9985–0.9999)

0.0044 (0.0035–0.0052)

XGBoost

0.9971 (0.9958–0.9982)

0.9997 (0.9995–0.9999)

0.9949 (0.9929–0.9964)

0.9764 (0.9693–0.983)

0.9538 (0.9374–0.9702)

0.999 (0.9983–0.9998)

0.0042 (0.0031–0.0054)

Random Forest

0.9966 (0.9951–0.9979)

0.9996 (0.9994–0.9998)

0.994 (0.9918–0.9957)

0.9745 (0.9671–0.9815)

0.9506 (0.9337–0.9676)

0.9984 (0.9974–0.9994)

0.0048 (0.004–0.0056)

Support Vector Machine

0.9935 (0.9904–0.9962)

0.999 (0.9983–0.9997)

0.9941 (0.992–0.9958)

0.9746 (0.9673–0.9815)

0.9506 (0.9337–0.9676)

0.9985 (0.9976–0.9995)

0.0047 (0.0036–0.0058)

Neural Network

0.993 (0.9885–0.9966)

0.9993 (0.9989–0.9997)

0.9934 (0.9912–0.9952)

0.972 (0.9644–0.9793)

0.9459 (0.9282–0.9636)

0.9982 (0.9972–0.9993)

0.0049 (0.0038–0.006)

kNN

0.9846 (0.979–0.9894)

0.9955 (0.9923–0.9988)

0.9864 (0.9833–0.989)

0.9324 (0.921–0.9433)

0.8662 (0.8396–0.8929)

0.9985 (0.9976–0.9995)

0.0095 (0.0081–0.011)

Logistic Regression

0.9791 (0.9648–0.9905)

0.9973 (0.9952–0.9994)

0.9915 (0.989–0.9935)

0.9681 (0.9602–0.9757)

0.9395 (0.9208–0.9581)

0.9968 (0.9954–0.9982)

0.0063 (0.0051–0.0077)

Regularized Linear Discriminal Analysis

0.9682 (0.9498–0.9856)

0.9977 (0.9968–0.9986)

0.9831 (0.9798–0.986)

0.9134 (0.9009–0.9259)

0.828 (0.7985–0.8575)

0.9989 (0.998–0.9997)

0.0138 (0.0118–0.016)

SpO2 - SO2 difference (matched)

0.9311 (0.911–0.9498)

0.9825 (0.9761–0.9888)

0.9871 (0.9841–0.9896)

0.9342 (0.9228–0.9451)

0.8694 (0.8431–0.8958)

0.999 (0.9983–0.9998)

0.0124 (0.0103–0.0145)